Reproducibility of dietary intakes of macronutrients, specific food groups,and dietary patterns in 211 050 adults in the UK Biobank study

Abstract

To detect modest associations of dietary intake with disease risk, observational studies need to be large and control for moderate measurement errors. Thereproducibility of dietary intakes of macronutrients, food groups and dietary patterns (vegetarian and Mediterranean) was assessed in adults in theUKBiobank study on up tofive occasions using a web-based 24-h dietary assessment (n211 050), and using short FFQ recorded at baseline (n502 655)and after 4 years (n20 346). When the means of two 24-h assessments were used, the intra-class correlation coefficients (ICC) for macronutrients variedfrom 0·63 for alcohol to 0·36 for polyunsaturated fat. The ICC for food groups also varied from 0·68 for fruit to 0·18 forfish. The ICC for the FFQ variedfrom 0·66 for meat and fruit to 0·48 for bread and cereals. The reproducibility was higher for vegetarian status (κ>0·80) than for the Mediterraneandietary pattern (ICC = 0·45). Overall, the reproducibility of pairs of 24-h dietary assessments and single FFQ used in the UK Biobank were comparablewith results of previous prospective studies using conventional methods. Analyses of diet–disease relationships need to correct for both measurement errorand within-person variability in dietary intake in order to reliably assess any such associations with disease in the UK Biobank.